method
active
method:activation-patching-along-manifold-pca-componentsActivation patching along manifold PCA components
Interpolating along PCA centroids to show causal effect.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Do Sparse Autoencoders Capture Concept Manifolds?introducesmentions
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- The low-dimensional geometric structure discovered in neural activation space; contrasted with linear/Euclidean geometry.
- Standard method in mechanistic interpretability that intervenes on activations; VPD flips this paradigm by patching parameters.
- Manifold fitted to representations in activation space.
- Substitute activations to observe effects.
- Low-dimensional manifold of residual activations.
- PCA on 171 emotion probe activations across all tokens to produce ordered linear combinations and test if lower PCs are more persistent
- Substitutes mean activation for count c to modify perceived counts.
- Reveals low-dimensional counting manifolds.